IBM 7094 Text to Speech History and Evolution

When Computers First Found Their Voice: The Story of the IBM 7094 and “Daisy Bell”

Long before artificial intelligence could effortlessly clone human voices or navigate nuanced text-to-speech conversations, machines were entirely mute. They crunched numbers, spat out long rows of punch cards, and hummed behind glass walls, but they made no sound resembling human language.

That changed in the early 1960s at Bell Laboratories, where a team of researchers achieved a monumental milestone: teaching a massive mainframe computer to speak and sing. At the heart of this acoustic breakthrough was powerhouse computing hardware—frequently associated with systems like the IBM 704 and its successor, the IBM 7094—marking the birth of computerized speech synthesis.

The Science of Synthetic Speech

In 1961, physicist John Larry Kelly Jr., alongside programmer Carol Lochbaum and musical collaborator Max Mathews, set out to prove that human speech could be entirely simulated through physics and mathematics rather than recorded.

Instead of playing back pre-recorded human audio fragments, the team modeled the human vocal tract mathematically. They simulated:

• The vocal cords: Generating a buzzing or periodic pulse sound.
• The vocal tract transitions: Shaping resonant frequencies (formants) to mimic the movement of the tongue, lips, and jaw to form distinct vowels and consonants.

Because the computer hardware of the era lacked the processing power to generate raw audio waveforms directly in software fast enough, the IBM mainframe controlled an external vocoder (voice encoder). The computer sent digital instructions that modulated the electronic filter circuits of the vocoder, essentially playing it like an electronic instrument to produce intelligible phonemes.

“Daisy, Daisy, Give Me Your Answer, Do”

To showcase this futuristic technology, the research team decided the computer needed to do more than just recite isolated words or dry sentences—it had to sing.

Max Mathews arranged the classic 1892 popular song “Daisy Bell” (also known as “A Bicycle Built for Two”) for the machine. Kelly and Lochbaum carefully hand-coded the pitch and timing parameters for every single syllable onto punch cards, feeding them into the mainframe.

When the IBM system booted up its program and began to perform, it produced a slow, metallic, slightly eerie, yet completely recognizable mechanical rendition of the chorus:

“Daisy, Daisy, give me your answer, do!
I’m half crazy, all for the love of you…”

Accompanied by computerized musical notes generated by Max Mathews’ MUSIC program, it was an astonishing cultural and technological marvel. For the first time in history, a machine had sung a song.

A Legendary Legacy in Science Fiction

The demonstration left a profound mark on everyone who witnessed it—most notably, the famed science fiction author Arthur C. Clarke.

Clarke happened to be visiting Bell Laboratories around the time of the demonstration and was thoroughly mesmerized by the singing computer. Years later, when writing 2001: A Space Odyssey (both the novel and the screenplay alongside director Stanley Kubrick), Clarke immortalized the moment.

In the film’s chilling climax, the rogue sentient supercomputer HAL 9000 is being systematically deactivated. As its higher functions fade away, HAL regresses to its digital childhood and sings the very same song the IBM mainframe had performed years earlier: “Daisy, Daisy, give me your answer, do…”

From Mainframe Curiosities to Modern Reality

Today, the raw audio recording of the IBM system’s rendition of “Daisy Bell” is preserved in the United States National Recording Registry as a cultural treasure. It serves as a reminder of how far speech technology has evolved.

What once required a room-sized mainframe computer, complex physical modeling, and painstaking manual punch-card coding now happens instantly inside pocket-sized smartphones and smart speakers. Yet, every time an artificial intelligence voice assistant speaks fluidly to us today, it stands on the shoulders of those pioneering researchers in 1961 who first taught an IBM machine how to carry a tune.